Predicting Endotracheal Intubation Needs in Neonatal Intensive Care Unit: A Multimodal Approach

Background and objectives: Neonatal intubation is a widely used procedure for high-risk neonates. However, it is unknown about the appropriate timing of intubation. Our study aims to use an artificial intelligence (AI) model to predict the need for endotracheal intubation for neonatal intensive care unit (NICU) patients. Methods: We retrospectively collected 432 patients (98 intubated, 334 nonintubated) admitted to the NICU of Chungbuk National University Hospital from July 2022 to July 2023. The dataset consists of prenatal factors, arterial blood gas analysis (ABGA), vital signs, and clinical risk index (CRIB) II. With the collected dataset, we propose a multimodal of MLP and LSTM validated by four-fold cross-validation. Results: The proposed model performed an accuracy of 0.8132 ± 0.04, sensitivity of 0.7619 ± 0.14, specificity of 0.8295 ± 0.09, and AUROC 0.8387 ± 0.02. We demonstrated the proposed model's better accuracy and AUROC than other models. Conclusions: We expect the proposed model to reduce the complications of endotracheal intubation in neonates and provide a safer environment for clinicians and patients.